You ask AI to draft a customer reply.
The draft arrives quickly. Then you check the details, remove a promise your business cannot make and rewrite the wording.
Was it quicker than writing the reply yourself?
The speed of the first draft cannot answer that.
This week, choose one small task and measure the time needed to reach an answer you would actually use.
1. CHOOSE A TASK YOU CAN CHECK
Start with a task you understand well enough to judge.
For example: draft a short reply to an invented enquiry using public information about your service.
Keep the boundaries from Issue 001: an approved business account, an approved task, approved information and a responsible human.
If any of those are unclear, stop and ask.
For this first test, use invented information. Keep real customer correspondence out of it.
2. DEFINE WHAT A GOOD ANSWER NEEDS
Write a short checklist before you start.
For our invented enquiry, the reply must:
Answer the customer’s question.
Use only the supplied service information.
Avoid unsupported prices, deadlines or promises.
Make the next step clear.
Here is an example instruction:
Draft a short reply to this invented enquiry: “Can you deliver a parcel tomorrow?” Use only the public service information supplied below. Do not invent availability, prices or delivery guarantees. If information is missing, identify what needs checking.
Add the public service information you have approved for the task.
The checklist gives you a consistent way to assess both your own reply and AI’s draft.
3. COUNT THE WHOLE JOB
Compare similar tasks against the same checklist.
Record the time spent writing normally. For the AI attempt, include preparing the instruction, reading the output, checking it and making corrections.
Invented example, not a measured result:
Method | Total time to an acceptable reply |
|---|---|
Write and check it yourself | 8 minutes |
Prepare the AI instruction | 2 minutes |
Generate and read the draft | 1 minute |
Check and correct the draft | 6 minutes |
AI attempt total | 9 minutes |
In this example, the quick draft produced no time saving.
The reverse could happen. Measure it rather than assuming either outcome.
THE TAKEAWAY
Measure the finished work.
A quicker result only counts if it meets the same quality requirements.
One attempt is a starting point. Repeat the comparison on a few similar, low-risk examples before deciding whether to keep using AI for that task. Note mistakes as well as minutes.
ONE ACTION
Test one approved task this week. Record the total time and whether each result passed your checklist.
Then choose KEEP, ADAPT or REJECT for that task.
Barbara
AI Made Human
P.S. If you cannot confidently check the answer, choose a simpler task for your first test.